♟ arsenm

There are some test crashes with this and I think I made the wrong guess for dmask behavior

That seems like a clang bug? I don't understand what the option is for, but isn't there some null streamer it's supposed to use?

I also don't understand why this would be necessary. How can the streamer not be available?

Should the name just be WQM since it seems to map directly to the WQM instruction?

AMDGPU: Implement hasBitPreservingFPLogic

AMDGPU: Look for src mods before fp_extend

AMDGPU: Implement isFPExtFoldable

DAG: Add opcode and source type to isFPExtFree

DAG: Add flags to dumps

In D36706#891897, @nemanjai wrote:

It would appear to me that the PPC-only portion of this is NFC since it'll still return true for any floating point destination type. So I would say that the PPC back end is fine with those (as long as the assert gets a message :)).
However, shouldn't this patch have a test case?

AMDGPU: Fix missing skipFunction calls

AMDGPU: Fix failure to select branch with optnone

AMDGPU: Fix incorrect selection of pseudo-branches

Right version of test

AMDGPU: Add read_exec_lo/hi builtins

AMDGPU: Fix missing declaration for __builtin_amdgcn_dispatch_ptr

Add source type and rename

OpenCL: Assume functions are convergent

Is this a port of what SC emits? I thought it emitted control flow for this.

AMDGPU: Set v2i32 any_extend to expand

Also update fsub case

AMDGPU: Add comment about clamps

AMDGPU: Do not fold clamp instructions when sources are different

AMDGPU: Fix not accounting for instruction size in bundles

Needs more comprehensive check lines. Just checking the instructions won't demonstrate that the extra instructions you're trying to avoid aren't there

ping, I think I ran into this same issue on another testcase

LGTM. It shouldn't be any different for the f32 vectors.

In D36706#843207, @RKSimon wrote:

Given its limited use I don't see why we shouldn't just update the existing isFPExtFree to take Opcode, SrcVT and DstVT - all of which may be useful.

AMDGPU: Remove global isGCN predicates

I want to replace this pass by always inserting the branches on execz, and have a new pass which optimizes out short jumps. Would that be easier than trying to analyze this?

AMDGPU: Fix potentially incorrectly matching check lines

Remove an unnecessary SubtargetPredicate block, fix a few more globally overwriting let Predicates = blocks

In D37894#873521, @uabelho wrote:
Pass DataLayout to isNoopCast when called from lint.

Question:
In FastISel::hasTrivialKill there is:
// No-op casts are trivially coalesced by fast-isel.
if (const auto *Cast = dyn_cast<CastInst>(I))
  if (Cast->isNoopCast(DL.getIntPtrType(Cast->getContext())) &&
Do you know if it is ok to change this isNoopCast call to Cast->isNoopCast(DL)?

If so, then I can clean up the patch further by removing the IntPtrTy versions completely.

arsenm (Matt Arsenault)
User

Projects

User Details

Recent Activity

Tue, Oct 17

Mon, Oct 16

Fri, Oct 13

Thu, Oct 12

Wed, Oct 11

Tue, Oct 10

Mon, Oct 9

Sat, Oct 7

Fri, Oct 6

Thu, Oct 5

Wed, Oct 4

Tue, Oct 3

Mon, Oct 2

arsenm (Matt Arsenault)User